nextpnr/mistral/pack.cc

517 lines
21 KiB
C++
Raw Permalink Normal View History

/*
* nextpnr -- Next Generation Place and Route
*
* Copyright (C) 2021 gatecat <gatecat@ds0.me>
*
* Permission to use, copy, modify, and/or distribute this software for any
* purpose with or without fee is hereby granted, provided that the above
* copyright notice and this permission notice appear in all copies.
*
* THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
* WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
* MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
* ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
* WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
* ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
* OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
*
*/
#include "design_utils.h"
#include "log.h"
#include "nextpnr.h"
#include "util.h"
NEXTPNR_NAMESPACE_BEGIN
namespace {
struct MistralPacker
{
MistralPacker(Context *ctx) : ctx(ctx) {};
Context *ctx;
NetInfo *gnd_net, *vcc_net;
void init_constant_nets()
{
CellInfo *gnd_drv = ctx->createCell(ctx->id("$PACKER_GND_DRV"), id_MISTRAL_CONST);
gnd_drv->params[id_LUT] = 0;
gnd_drv->addOutput(id_Q);
CellInfo *vcc_drv = ctx->createCell(ctx->id("$PACKER_VCC_DRV"), id_MISTRAL_CONST);
vcc_drv->params[id_LUT] = 1;
vcc_drv->addOutput(id_Q);
gnd_net = ctx->createNet(ctx->id("$PACKER_GND_NET"));
vcc_net = ctx->createNet(ctx->id("$PACKER_VCC_NET"));
gnd_drv->connectPort(id_Q, gnd_net);
vcc_drv->connectPort(id_Q, vcc_net);
}
CellPinState get_pin_needed_muxval(CellInfo *cell, IdString port)
{
NetInfo *net = cell->getPort(port);
if (net == nullptr || net->driver.cell == nullptr) {
// Pin is disconnected
// If a mux value exists already, honour it
CellPinState exist_mux = cell->get_pin_state(port);
if (exist_mux != PIN_SIG)
return exist_mux;
// Otherwise, look up the default value and use that
CellPinStyle pin_style = ctx->get_cell_pin_style(cell, port);
if ((pin_style & PINDEF_MASK) == PINDEF_0)
return PIN_0;
else if ((pin_style & PINDEF_MASK) == PINDEF_1)
return PIN_1;
else
return PIN_SIG;
}
// Look to see if the driver is an inverter or constant
IdString drv_type = net->driver.cell->type;
if (drv_type == id_MISTRAL_NOT)
return PIN_INV;
else if (drv_type == id_GND)
return PIN_0;
else if (drv_type == id_VCC)
return PIN_1;
else
return PIN_SIG;
}
void uninvert_port(CellInfo *cell, IdString port)
{
// Rewire a port so it is driven by the input to an inverter
NetInfo *net = cell->getPort(port);
NPNR_ASSERT(net != nullptr && net->driver.cell != nullptr && net->driver.cell->type == id_MISTRAL_NOT);
CellInfo *inv = net->driver.cell;
cell->disconnectPort(port);
NetInfo *inv_a = inv->getPort(id_A);
if (inv_a != nullptr) {
cell->connectPort(port, inv_a);
}
}
void process_inv_constants(CellInfo *cell)
{
// TODO: we might need to create missing inputs here in some cases so we can tie them to the correct constant?
// Fold inverters and constants into a cell
for (auto &port : cell->ports) {
// Iterate over all inputs
if (port.second.type != PORT_IN)
continue;
IdString port_name = port.first;
CellPinState req_mux = get_pin_needed_muxval(cell, port_name);
if (req_mux == PIN_SIG) {
// No special setting required, ignore
continue;
}
CellPinStyle pin_style = ctx->get_cell_pin_style(cell, port_name);
if (req_mux == PIN_INV) {
// Pin is inverted. If there is a hard inverter; then use it
if (pin_style & PINOPT_INV) {
uninvert_port(cell, port_name);
cell->pin_data[port_name].state = PIN_INV;
}
} else if (req_mux == PIN_0 || req_mux == PIN_1) {
// Pin is tied to a constant
// If there is a hard constant option; use it
if ((pin_style & int(req_mux)) == req_mux) {
cell->disconnectPort(port_name);
cell->pin_data[port_name].state = req_mux;
} else {
cell->disconnectPort(port_name);
// There is no hard constant, we need to connect it to the relevant soft-constant net
cell->connectPort(port_name, (req_mux == PIN_1) ? vcc_net : gnd_net);
}
}
}
}
void trim_design()
{
// Remove unused inverters and high/low drivers
std::vector<IdString> trim_cells;
std::vector<IdString> trim_nets;
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->type != id_MISTRAL_NOT && ci->type != id_GND && ci->type != id_VCC)
continue;
IdString port = (ci->type == id_MISTRAL_NOT) ? id_Q : id_Y;
NetInfo *out = ci->getPort(port);
if (out == nullptr) {
trim_cells.push_back(ci->name);
continue;
}
if (!out->users.empty())
continue;
ci->disconnectPort(id_A);
trim_cells.push_back(ci->name);
trim_nets.push_back(out->name);
}
for (IdString rem_net : trim_nets)
ctx->nets.erase(rem_net);
for (IdString rem_cell : trim_cells)
ctx->cells.erase(rem_cell);
}
void pack_constants()
{
// Iterate through cells
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
// Skip certain cells at this point
if (ci->type != id_MISTRAL_NOT && ci->type != id_GND && ci->type != id_VCC)
process_inv_constants(ci);
}
// Special case - SDATA can only be trimmed if SLOAD is low
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->type != id_MISTRAL_FF)
continue;
if (ci->get_pin_state(id_SLOAD) != PIN_0)
continue;
ci->disconnectPort(id_SDATA);
}
// Remove superfluous inverters and constant drivers
trim_design();
}
void prepare_io()
{
// Find the actual IO buffer corresponding to a port; and copy attributes across to it
// Note that this relies on Yosys to do IO buffer inference, to avoid tristate issues once we get to synthesised
// JSON. In all cases the nextpnr-inserted IO buffers are removed as redundant.
for (auto &port : ctx->ports) {
if (!ctx->cells.count(port.first))
log_error("Port '%s' doesn't seem to have a corresponding top level IO\n", ctx->nameOf(port.first));
CellInfo *ci = ctx->cells.at(port.first).get();
PortRef top_port;
top_port.cell = nullptr;
bool is_npnr_iob = false;
if (ci->type == ctx->id("$nextpnr_ibuf") || ci->type == ctx->id("$nextpnr_iobuf")) {
// Might have an input buffer (IB etc) connected to it
is_npnr_iob = true;
NetInfo *o = ci->getPort(id_O);
if (o == nullptr)
;
else if (o->users.entries() > 1)
log_error("Top level pin '%s' has multiple input buffers\n", ctx->nameOf(port.first));
else if (o->users.entries() == 1)
top_port = *o->users.begin();
}
if (ci->type == ctx->id("$nextpnr_obuf") || ci->type == ctx->id("$nextpnr_iobuf")) {
// Might have an output buffer (OB etc) connected to it
is_npnr_iob = true;
NetInfo *i = ci->getPort(id_I);
if (i != nullptr && i->driver.cell != nullptr) {
if (top_port.cell != nullptr)
log_error("Top level pin '%s' has multiple input/output buffers\n", ctx->nameOf(port.first));
top_port = i->driver;
}
// Edge case of a bidirectional buffer driving an output pin
if (i->users.entries() > 2) {
log_error("Top level pin '%s' has illegal buffer configuration\n", ctx->nameOf(port.first));
} else if (i->users.entries() == 2) {
if (top_port.cell != nullptr)
log_error("Top level pin '%s' has illegal buffer configuration\n", ctx->nameOf(port.first));
for (auto &usr : i->users) {
if (usr.cell->type == ctx->id("$nextpnr_obuf") || usr.cell->type == ctx->id("$nextpnr_iobuf"))
continue;
top_port = usr;
break;
}
}
}
if (!is_npnr_iob)
log_error("Port '%s' doesn't seem to have a corresponding top level IO (internal cell type mismatch)\n",
ctx->nameOf(port.first));
if (top_port.cell == nullptr) {
log_info("Trimming port '%s' as it is unused.\n", ctx->nameOf(port.first));
} else {
// Copy attributes to real IO buffer
if (ctx->io_attr.count(port.first)) {
for (auto &kv : ctx->io_attr.at(port.first)) {
top_port.cell->attrs[kv.first] = kv.second;
}
}
// Make sure that top level net is set correctly
port.second.net = top_port.cell->ports.at(top_port.port).net;
}
// Now remove the nextpnr-inserted buffer
ci->disconnectPort(id_I);
ci->disconnectPort(id_O);
ctx->cells.erase(port.first);
}
}
void pack_io()
{
// Step 0: deal with top level inserted IO buffers
prepare_io();
// Stage 1: apply constraints
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
// Iterate through all IO buffer primitives
if (!ctx->is_io_cell(ci->type))
continue;
// We need all IO constrained at the moment, unconstrained IO are rare enough not to care
if (!ci->attrs.count(id_LOC))
log_error("Found unconstrained IO '%s', these are currently unsupported\n", ctx->nameOf(ci));
// Convert package pin constraint to bel constraint
std::string loc = ci->attrs.at(id_LOC).as_string();
if (loc.compare(0, 4, "PIN_") != 0)
log_error("Expecting PIN_-prefixed pin for IO '%s', got '%s'\n", ctx->nameOf(ci), loc.c_str());
auto pin_info = ctx->cyclonev->pin_find_name(loc.substr(4));
if (pin_info == nullptr)
log_error("IO '%s' is constrained to invalid pin '%s'\n", ctx->nameOf(ci), loc.c_str());
BelId bel = ctx->get_io_pin_bel(pin_info);
if (bel == BelId()) {
log_error("IO '%s' is constrained to pin %s which is not a supported IO pin.\n", ctx->nameOf(ci),
loc.c_str());
} else {
log_info("Constraining IO '%s' to pin %s (bel %s)\n", ctx->nameOf(ci), loc.c_str(),
ctx->nameOfBel(bel));
ctx->bindBel(bel, ci, STRENGTH_LOCKED);
}
}
}
void constrain_carries()
{
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->type != id_MISTRAL_ALUT_ARITH)
continue;
const NetInfo *cin = ci->getPort(id_CI);
if (cin != nullptr && cin->driver.cell != nullptr)
continue; // not the start of a chain
std::vector<CellInfo *> chain;
CellInfo *cursor = ci;
while (true) {
chain.push_back(cursor);
const NetInfo *co = cursor->getPort(id_CO);
if (co == nullptr || co->users.empty())
break;
if (co->users.entries() > 1)
log_error("Carry net %s has more than one sink!\n", ctx->nameOf(co));
auto &usr = *co->users.begin();
if (usr.port != id_CI)
log_error("Carry net %s drives port %s, expected CI\n", ctx->nameOf(co), ctx->nameOf(usr.port));
cursor = usr.cell;
}
chain.at(0)->constr_abs_z = true;
chain.at(0)->constr_z = 0;
chain.at(0)->cluster = chain.at(0)->name;
for (int i = 1; i < int(chain.size()); i++) {
chain.at(i)->constr_x = 0;
chain.at(i)->constr_y = -(i / 20);
// 2 COMB, 4 FF per ALM
chain.at(i)->constr_z = ((i / 2) % 10) * 6 + (i % 2);
chain.at(i)->constr_abs_z = true;
chain.at(i)->cluster = chain.at(0)->name;
chain.at(0)->constr_children.push_back(chain.at(i));
}
if (ctx->debug) {
log_info("Chain: \n");
for (int i = 0; i < int(chain.size()); i++) {
auto &c = chain.at(i);
log_info(" i=%d cell=%s dy=%d z=%d ci=%s co=%s\n", i, ctx->nameOf(c), c->constr_y, c->constr_z,
ctx->nameOf(c->getPort(id_CI)), ctx->nameOf(c->getPort(id_CO)));
}
}
}
// Check we reached all the cells in the above pass
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->type != id_MISTRAL_ALUT_ARITH)
continue;
if (ci->cluster == ClusterId())
log_error("Failed to include arith cell '%s' in any chain (CI=%s)\n", ctx->nameOf(ci),
ctx->nameOf(ci->getPort(id_CI)));
}
}
void constrain_lutram()
{
// We form clusters based on both read and write address; as both being the same makes it more likely these
// cells should be packed together, too.
// This makes things easier for the placement legaliser to deal with RAM in LAB-compatible blocks without
// over-constraining things
idict<dict<IdString, IdString>> mlab_keys;
std::vector<std::vector<CellInfo *>> mlab_groups;
for (auto &cell : ctx->cells) {
CellInfo *ci = cell.second.get();
if (ci->type != id_MISTRAL_MLAB)
continue;
auto key = ctx->get_mlab_key(ci, true);
int key_idx = mlab_keys(key);
if (key_idx >= int(mlab_groups.size()))
mlab_groups.resize(key_idx + 1);
mlab_groups.at(key_idx).push_back(ci);
}
// Combine into clusters
size_t cluster_size = 20;
for (auto &group : mlab_groups) {
for (size_t i = 0; i < group.size(); i++) {
CellInfo *ci = group.at(i);
CellInfo *base = group.at((i / cluster_size) * cluster_size);
int cell_index = int(i) % cluster_size;
int alm = cell_index / 2;
int alm_cell = cell_index % 2;
ci->cluster = base->name;
ci->constr_abs_z = true;
ci->constr_z = alm * 6 + alm_cell;
if (cell_index != 0) {
// Not the root of a cluster
base->constr_children.push_back(ci);
ci->constr_x = 0;
ci->constr_y = 0;
}
}
}
}
2021-12-22 05:34:28 +08:00
void setup_m10ks()
{
for (auto &cell : ctx->cells) {
2021-12-22 05:34:28 +08:00
CellInfo *ci = cell.second.get();
if (ci->type != id_MISTRAL_M10K)
continue;
auto abits = ci->params.at(id_CFG_ABITS).as_int64();
auto dbits = ci->params.at(id_CFG_DBITS).as_int64();
2021-12-22 05:34:28 +08:00
NPNR_ASSERT(abits >= 7 && abits <= 13);
NPNR_ASSERT(dbits == 1 || dbits == 2 || dbits == 5 || dbits == 10 || dbits == 20 || dbits == 40);
NPNR_ASSERT((1 << abits) * dbits <= 10240);
log_info("Setting up %ld-bit address, %ld-bit data M10K for %s.\n", abits, dbits,
ci->name.str(ctx).c_str());
2021-12-22 05:34:28 +08:00
// Quartus doesn't seem to generate ADDRSTALL[AB], BYTEENABLE[AB][01].
// It *does* generate ACLR[01] but leaves them unconnected if unused.
// Enables.
// RDEN[1] is left unconnected.
if (dbits == 40)
ci->pin_data[ctx->id("A1EN")].bel_pins = {ctx->id("WREN[0]")};
else
ci->pin_data[ctx->id("A1EN")].bel_pins = {ctx->id("WREN[1]")};
2022-01-18 20:15:01 +08:00
ci->pin_data[ctx->id("B1EN")].bel_pins = {ctx->id("RDEN[0]")};
2021-12-22 05:34:28 +08:00
// Clocks.
ci->pin_data[ctx->id("CLK1")].bel_pins = {ctx->id("CLKIN[0]")};
// Enables left unconnected.
// Address lines.
// One could remove the std::max here and the `- bit_offset`s here,
// because they would cancel out, but I think this way is less confusing.
int addr_offset = std::max(12 - std::max(abits, dbits == 40 ? 8L : 9L), 0L);
int bit_offset = (abits == 13);
2021-12-22 05:34:28 +08:00
if (abits == 13) {
ci->pin_data[ctx->id("A1ADDR[0]")].bel_pins = {ctx->id("DATAAIN[4]")};
ci->pin_data[ctx->id("B1ADDR[0]")].bel_pins = {ctx->id("DATABIN[19]")};
}
for (int bit = bit_offset; bit < abits; bit++) {
ci->pin_data[ctx->idf("A1ADDR[%d]", bit)].bel_pins = {
ctx->idf("ADDRA[%d]", bit + addr_offset - bit_offset)};
ci->pin_data[ctx->idf("B1ADDR[%d]", bit)].bel_pins = {
ctx->idf("ADDRB[%d]", bit + addr_offset - bit_offset)};
2021-12-22 05:34:28 +08:00
}
// Data lines
std::vector<int> offsets;
offsets.push_back(0);
if (abits >= 10 && dbits <= 10) {
offsets.push_back(10);
}
if (abits >= 11 && dbits <= 5) {
offsets.push_back(5);
offsets.push_back(15);
}
if (abits >= 12 && dbits <= 2) {
offsets.push_back(2);
offsets.push_back(7);
offsets.push_back(12);
offsets.push_back(17);
}
if (abits == 13 && dbits == 1) {
offsets.push_back(1);
offsets.push_back(3);
offsets.push_back(6);
offsets.push_back(8);
offsets.push_back(11);
offsets.push_back(13);
offsets.push_back(16);
offsets.push_back(18);
}
// In this corner case the pin name does not have indexing
// because it's a single bit wide...
if (abits == 13 && dbits == 1) {
for (int offset : offsets)
ci->pin_data[ctx->idf("A1DATA")].bel_pins.push_back(ctx->idf("DATAAIN[%d]", offset));
ci->pin_data[ctx->idf("B1DATA")].bel_pins = {ctx->idf("DATABOUT[0]")};
continue;
}
// 40-bit data mode causes some headaches...
bit_offset = dbits == 40 ? 20 : 0;
// Write port
for (int bit = 0; bit < std::min(dbits, 20L); bit++)
for (int offset : offsets)
ci->pin_data[ctx->idf("A1DATA[%d]", bit)].bel_pins.push_back(ctx->idf("DATAAIN[%d]", bit + offset));
2021-12-22 05:34:28 +08:00
if (dbits == 40)
for (int bit = bit_offset; bit < dbits; bit++)
ci->pin_data[ctx->idf("A1DATA[%d]", bit)].bel_pins.push_back(
ctx->idf("DATABIN[%d]", bit - bit_offset));
// Read port
if (dbits == 40)
for (int bit = 0; bit < 20; bit++)
ci->pin_data[ctx->idf("B1DATA[%d]", bit)].bel_pins = {ctx->idf("DATAAOUT[%d]", bit)};
for (int bit = bit_offset; bit < dbits; bit++)
ci->pin_data[ctx->idf("B1DATA[%d]", bit)].bel_pins = {ctx->idf("DATABOUT[%d]", bit - bit_offset)};
2021-12-22 05:34:28 +08:00
}
}
void run()
{
init_constant_nets();
pack_constants();
pack_io();
constrain_carries();
constrain_lutram();
setup_m10ks();
}
};
}; // namespace
bool Arch::pack()
{
MistralPacker packer(getCtx());
packer.run();
assignArchInfo();
return true;
}
NEXTPNR_NAMESPACE_END